You are on page 1of 18

What Is ModelArts?

ModelArts is a one-stop development platform for AI developers. With data


preprocessing, semi-automated data labeling, distributed training, automated
model building, and on-demand model deployment on the device, edge, and
cloud, ModelArts helps AI developers quickly build models and efficiently
manage the AI development lifecycle.

ModelArts covers all stages of AI development, including data processing and


model training and deployment.
The underlying technologies of ModelArts support various heterogeneous
computing resources, allowing developers to flexibly select and use resources.

In addition, ModelArts supports popular open source AI development frameworks


such as TensorFlow and MXNet. Developers can also use self-developed
algorithm frameworks to match their usage habits.

ModelArts aims to simplify AI development. ModelArts is suitable for AI


developers with varying levels of development experience. Service developers
can use ExeML to quickly build AI applications without coding. Beginners can
directly use built-in algorithms to build AI applications.

AI engineers can use multiple development environments to quickly compile code


for modeling and application development.
ModelArts Architecture and Application

ModelArts supports the entire development process, including data processing,


and model training, management, and deployment. ModelArts supports various
AI application scenarios, such as image classification, video analysis, speech
recognition, product recommendation, and exception detection, and object
detection.
Product Advantages
● One-stop platform
The out-of-the-box and full-lifecycle AI development platform provides one-stop
data processing, and development, training, management, and deployment of
models.

● Easy to use
– Multiple built-in models provided and free use of open source models
– Automatic optimization of hyper-parameters
– Code-free development and simplified operations
– One-click deployment of models to the cloud, edge, and devices

● High performance
– The self-developed MoXing deep learning framework accelerates algorithm
development and training.
– Optimized GPU utilization accelerates real-time inference.
– Models running on Ascend AI chips achieve more efficient inference.

● Flexible
– Popular open source frameworks available, such as TensorFlow, Spark_MLlib,
MXNet, Caffe, PyTorch, XGBoost-Sklearn, and MindSpore
– Popular GPUs and self-developed Ascend chips available
– Exclusive use of dedicated resources
– Custom images for custom frameworks and operators

Using ModelArts for the First Time –Applying ModelArts

If you are a first-time user, the following information will help you get familiar with
ModelArts:
● Basic concepts
Basic Knowledge describes the basic concepts of ModelArts, including the
basic process and concepts of AI development, and specific concepts and
functions of ModelArts.
● Getting started
The Getting Started document provides detailed operation guides to walk you
through model building on ModelArts.
● Best practices
ModelArts supports multiple open source engines and provides extensive use
cases based on the engines and functions. You can build and deploy models by
referring to Best Practices.

● Other functions and operation guides

– If you are a service developer, you can use ExeML to quickly build models
without coding.

– If you are a beginner, you can use common AI algorithms to quickly build
models without coding. ModelArts provides built-in algorithms based on the
common AI engines.

– If you are an AI engineer, you can manage the AI development lifecycle,


including data management, and model development, training, management, and
deployment.

– If you are a developer and want to use ModelArts APIs or SDKs for AI
development.
ModelArts Functions
AI engineers face challenges in the installation and configuration of various AI
tools, data preparation, and model training. To address these challenges, the
one-stop AI development platform ModelArts is provided.

The platform integrates data preparation, algorithm development, model training,


and model deployment into the production environment, allowing AI engineers to
perform one-stop AI development.

Figure showing the Function overview


Functions
 Data Management
ModelArts manages data preparation, such as collection, filtering, and labeling,
and dataset versions, especially for deep learning datasets.
 Rapid and Simplified Model Training
Huawei's MoXing deep learning framework enables high-performance distributed
training. To accelerate model development, it uses automatic hyperparameter
tuning and standalone and distributed training.
 Model Deployment
ModelArts deploys models in various production environments such as devices,
the edge, and the cloud, and supports online and batch inference jobs.
 ExeML
ModelArts supports code-free modeling and auto learning with image
classification, object detection, and predictive analytics.
 Visualized Workflow
ModelArts works with Graph Engine Service (GES) to manage and visualize the
lifecycle of AI development workflows, implementing data and model lineage.
 AI Marketplace
ModelArts supports common models and datasets, and internal or public sharing
of enterprise models in the marketplace.

ModelArts has the following features:

● Data governance
Manages data preparation, such as data filtering and labeling, and dataset
versions.
● Rapid and simplified model training
Enables high-performance distributed training and simplifies coding with the
self-developed MoXing deep learning framework.
● Cloud-edge-device synergy
Deploys models in various production environments such as devices, the edge,
and the cloud, and supports real-time and batch inference.
● Auto learning
Enables model building without coding and supports image classification, object
detection, and predictive analytics.
Introduction to the AI Development Lifecycle
What Is AI
Artificial intelligence (AI) is a technology capable of simulating human cognition
through machines. The core capability of AI is to make a judgment or prediction
based on a given input.

What Is the Purpose of AI Development


AI development aims to centrally process and extract information from volumes
of data to summarize internal patterns of the study objects.
Massive volumes of collected data are computed, analyzed, summarized, and
organized by using appropriate statistics, machine learning, and deep learning
methods to maximize data value.

Basic Process of AI Development


The basic process of AI development includes the following steps: determining
an objective, preparing data, and training, evaluating, and deploying a model.

Figure showing the AI development process

Step 1 Determine an objective.


Before starting AI development, determine what to analyze. What problems do
you want to solve? What is the business goal? Sort out the AI development
framework and ideas based on the business understanding. For example, image
classification and object detection. Different projects have different requirements
for data and AI development methods.

Step 2 Prepare data.

Data preparation refers to data collection and preprocessing.


Data preparation is the basis of AI development. When you collect and integrate
related data based on the determined objective, the most important thing is to
ensure the authenticity and reliability of the obtained data. Typically, you cannot
collect all the data at the same time. In the data labeling phase, you may find that
some data sources are missing and then you may need to repeatedly adjust and
optimize the data.
Step 3 Train a model.
Modeling involves analyzing the prepared data to find the causality, internal
relationships, and regular patterns, thereby providing references for commercial
decision making. After model training, usually one or more machine learning or
deep learning models are generated. These models can be applied to new data
to obtain predictions and evaluation results.
A large number of developers develop and train models required by relevant
services based on popular AI engines, such as TensorFlow, Spark_MLlib,
MXNet, Caffe, PyTorch, XGBoost-Sklearn, and MindSpore.

Step 4 Evaluate the model.


A model generated by training needs to be evaluated. Typically, you cannot
obtain a satisfactory model after the first evaluation, and may need to repeatedly
adjust algorithm parameters and data to further optimize the model.
Some common metrics, such as the accuracy, recall, and area under the curve
(AUC), help you effectively evaluate and obtain a satisfactory model.

Step 5 Deploy the model.


Model development and training are based on existing data (which may be test
data). After a satisfactory model is obtained, the model needs to be formally
applied to actual data or newly generated data for prediction, evaluation, and
visualization. The findings can then be reported to decision makers in an intuitive
way, helping them develop the right business strategies.

Basic Concepts of AI Development


Machine learning is classified into supervised, unsupervised, and reinforcement
learning.
● Supervised learning uses labeled samples to adjust the parameters of
classifiers to achieve the required performance. It can be considered as learning
with a teacher. Common supervised learning includes regression and
classification.
● Unsupervised learning is used to find hidden structures in unlabeled data.
Clustering is a form of unsupervised learning.
● Reinforcement learning is an area of machine learning concerned with how
software agents ought to take actions in an environment so as to maximize some
notion of cumulative reward.
Common Concepts of ModelArts
ExeML
ExeML is the process of automating model design, parameter tuning, and model
training, model compression, and model deployment with the labeled data. The
process is code-free and does not require developers to have experience in
model development. A model can be built in three steps: labeling data, training a
model, and deploying the model.
Device-Edge-Cloud
Device-Edge-Cloud indicates devices, intelligent edge nodes, and the public
cloud.
Inference
Inference is the process of deriving a new judgment from a known judgment
according to a certain strategy. In AI, machines simulate human intelligence, and
complete inference based on neural networks.
Real-Time Inference
Real-time inference specifies a web service that provides an inference result for
each inference request.
Batch Inference
Batch inference specifies a batch job that processes batch data for inference.
Ascend Chip
The Ascend chips are a series of Huawei-developed AI chips with high
computing performance and low power consumption.
Modeling Patterns /Highlights
Modeling from Scratch

Intended For

Developers with AI application requirements but no AI development capabilities.


They are unfamiliar with AI development frameworks and cannot develop models
themselves.

Highlights

ModelArts provides ExeML to automate model design, parameter tuning and


training, and model compression and deployment with the labeled data. The
process is code-free and does not require experience with model development,
allowing developers to start from scratch.

Application Scenarios

You can use ExeML to quickly create image classification, object detection, and
predictive analytics models. New models are coming soon.
Quick Modeling

Intended For

AI beginners with basic knowledge but limited AI development capabilities. They


are able to use common AI development frameworks and open source tools to
create simple models.

Highlights

ModelArts offers built-in pre-trained algorithms. Without any coding, you can
upload your own service data, select a desired built-in algorithm, retrain the
algorithm to create a model, and deploy the model as a service.

Application Scenarios

ModelArts provides three types of algorithms: image classification, object


detection, and image segmentation. You can create a training job based on the
actual application requirements to obtain a desired model. More built-in models are being
added.
Standard Modeling

Intended For

AI engineers and experts with deep AI development capabilities, years of AI


development experience, and extensive model development and optimization
experience.

Highlights

ModelArts integrates Jupyter Notebook. You can create a development


environment, compile and debug the model training code, and use the compiled
code to create a training job to train and deploy a model. ModelArts supports
version management of datasets, training jobs, and models. It also provides
traceback diagrams of datasets, training jobs, models, and services to visualize
AI development workflows. This helps you easily manage AI development and
improve AI development efficiency.

Application Scenarios
ModelArts allows you to customize and deploy deep learning and conventional
machine learning models.

ModelArt Product Advantages

 One-Stop Platform

The out-of-the-box and full-lifecycle AI development platform provides one-stop data


processing, model development, training, management, and deployment.
 Easy to Use

Various built-in open source models and automatic hyperparameter tuning help you
start model training from scratch. Models can be deployed on the device, edge, and
cloud with just one click.
 Excellent Performance

The Huawei-developed MoXing framework delivers high-performance algorithm


development and training. GPU utilization is optimized for online inference. Huawei
Ascend chips significantly accelerate inference.
 High Flexibility

ModelArts supports multiple mainstream open source frameworks, such as TensorFlow


and Apache Spark MLlib, mainstream GPUs, and the Huawei-developed Ascend AI
chips. Exclusive use of resources and custom images ensure flexible development
experience.

How Do I Access ModelArts


You can access ModelArts through the web-based management console or by
using HTTPS-based application programming interfaces (APIs).

● Using the Management Console

ModelArts features a simple and easy-to-use management console, and provides


a host of functions including ExeML, data management, development
environment, model training, model management, and service deployment.

You can complete end-to-end AI development on the management console.

To use the ModelArts management console, you need to register with HUAWEI
CLOUD first. If you have registered with HUAWEI CLOUD, choose Enterprise
Intelligence > ModelArts on the official website and log in to the management
console.

● Using SDKs
If you want to integrate ModelArts into a third-party system for secondary
development, call SDKs to complete the development.

ModelArts SDKs encapsulate RESTful APIs provided by ModelArts to simplify


secondary development.

In addition, you can directly call the ModelArts SDKs when writing code in a
notebook on the management console.

● Using APIs
If you want to integrate ModelArts into a third-party system for secondary
development, use APIs to access ModelArts.

How to use ModelArts


ModelArts is a one-stop AI development platform provided by HUAWEI CLOUD.
It integrates the Ascend AI Processor resource pool. 
ModelArts is easy to use for users with different experience.
 For service developers without AI development experience, you can use
ExeML of ModelArts to build AI models without coding.
 For developers who are familiar with code compilation, debugging, and
common AI engines, ModelArts provides online code compiling
environments as well as AI development lifecycle that covers data
preparation, model training, model management, and service deployment,
helping the developers build models efficiently and quickly.

ExeML

ExeML is a customized code-free model development tool that helps users start
AI application development from scratch with high flexibility. ExeML automates
model design, parameter tuning and training, and model compression and
deployment with the labeled data. Developers do not need to develop basic and
encoding capabilities, but only to upload data and complete model training and
deployment as prompted by ExeML.

AI Development Lifecycle

 The AI development lifecycle on ModelArts takes developers' habits into


consideration and provides a variety of engines and scenarios for
developers to choose. The following describes the entire process from
data preparation to service development using ModelArts.

Figure showing Process of using ModelArts


Table 1 Process description

Table 1 Process description


Task Sub Task Description Reference
Data Create a dataset in
preparation Creating a ModelArts to manage
Creating a Dataset
dataset and preprocess your
business data.
Label and preprocess
the data in your dataset
based on the business
logic to facilitate
Labeling data Labeling Data
subsequent training.
Data labeling affects the
model training
performance.
After labeling data,
publish the database to
Publishing a
generate a dataset Publishing a Dataset
dataset
version that can be used
for model training.
Development Creating a Create a notebook Creating a Notebook
Table 1 Process description
Task Sub Task Description Reference
instance as the
notebook
development Instance
instance
environment.
Opening a Notebook
Compile code in an
Compiling Instance
existing notebook to
code Common Operations
directly build a model.
on Jupyter Notebook
Export the compiled
training script as
Exporting a .py file for subsequent Using the Convert to
the .py file operations, such as Python File Function
model training and
management.
Model Create a training job,
training upload and use the
compiled training script.
Creating a
After the training is Introduction
training job
complete, a model is
generated and stored in
OBS.
Create a visualization
job (TensorBoard type)
to view the model
(Optional) training process, learn
Creating a about the model, and Managing
visualization adjust and optimize the Visualization Jobs
job model. Currently,
visualization jobs only
support the MXNet and
TensorFlow engines.
Model Compiling Following the model Model Package
management inference package specifications Specifications
code and provided by ModelArts,
configuration compile inference code
files and configuration files
for your model, and
save the inference code
and configuration files to
the training output
location.
Table 1 Process description
Task Sub Task Description Reference
Import the training
Importing a model to ModelArts to Model Management
model facilitate service Overview
deployment.
Model  Deploying a
deployment Model as a Real-
Deploying a Deploy a model as a
Time Service
model as a real-time or batch
 Deploying a
service service.
Model as a Batch
Service
 Accessing a
After the service is Real-Time
deployed, access the Service (Token-
Accessing real-time service, or based
the service view the prediction Authentication)
result of the batch  Viewing the
service. Batch Service
Prediction Result

Benefits of ModelArts as in the Open Telekom Cloud

 Everything from a single source


From data preparation to training, to provision of the application – with us,
you get everything from a single source.

 Simple and versatile application


ModelArts can be used even without any programming experience and
supports common open-source frameworks.

 Performance
Requirement-based, powerful computing resources for quick training
(CPU, GPU, from Q3/2021 also NPU).

You might also like